Pick between CBO and ABO budget control with one number, conversions per ad set per week, instead of folklore. When each wins for testing, scaling, and small accounts.
Tuesday, 09:40. You check the scaled campaign you moved to CBO last week and Meta has already pushed 78% of today's budget into one ad set. The one that picked up three cheap conversions overnight. The other four ad sets, including the audience that carried last month, are getting scraps.
Now you have a decision every Meta buyer eventually faces: trust the machine or take the wheel back. I ran Meta budgets at €150k a month and I have switched camps on this question more than once. What finally settled it for me was realizing it is the wrong question. CBO vs ABO is not a loyalty test. It is an arithmetic problem, and the arithmetic is one number: conversions per ad set per week.
The takeaways
Because campaign budget optimization allocates spend toward whichever ad set its model predicts will convert cheapest, and early in an ad set's life that prediction rests on a handful of conversions. Three conversions is a coin that landed heads three times. With small samples, the early leader is usually the lucky one, and CBO funds the luck.
This is regression to the mean doing what it always does: extreme early results drift back toward average once more data arrives. I wrote a whole piece on how a lucky creative gets crowned a winner in manual testing. CBO automates the crowning. The budget moves before the evidence exists.
None of that makes CBO bad. Fed enough conversions, the same algorithm reallocates faster and with fewer ego attachments than any human rebalancing in Ads Manager at 11pm. The question is whether your account produces enough signal to feed it.
A workable bar already exists: Meta's documentation puts learning-phase exit at about 50 conversion events per ad set within 7 days. Use that as your signal-density test. If the ad sets inside a campaign cannot each plausibly reach 50 weekly conversions, CBO is choosing between options it cannot tell apart.
Run the arithmetic on your own account. A $200/day campaign with 5 ad sets and a $50 CPA produces about 4 conversions a day, 28 a week, across the whole campaign. No ad set gets near 50. CBO here is reshuffling noise, and you will watch it chase a different "winner" every few days.
Now the other end: $5,000/day across 3 proven ad sets at a $25 CPA is 200 conversions a day. Each ad set can clear the learning bar in two days. At that density the predictions have something to stand on, and consolidating budget into one CBO campaign usually beats manually nudging three separate caps.
ABO guarantees spend per hypothesis. If your test plan says every new creative concept gets $40 before judgment, only ad set budgets make that promise hold. Put the same test inside a CBO and the algorithm will starve half the contenders by lunchtime, exactly as in the Tuesday scenario above, and you end the week with five creatives you still know nothing about.
That guarantee has a price. With ABO you knowingly fund losers, because the spend on a losing ad set bought you the information that it loses. Testing is an information purchase. Pay for clean reads with ABO, then move what survives into CBO once it has the conversion volume to defend itself. The common version of this is duplicating the winning ad into a CBO scaling campaign via its post ID while the ABO original keeps running.
Less than they look. Figures like "27% lower cost per result with CBO" and "8 to 15% better CPA above $5k/day" circulate in tool-vendor blog posts, and none that I have found publishes sample sizes, account-size controls, or test design. Nobody outside Meta has the cross-account data to settle the argument, and Meta has an obvious interest in you handing over allocation.
The test you can own: duplicate your proven ad sets into a CBO campaign, keep the ABO originals live at their current caps, and compare CPA over two weeks. One account, your data, a result you can act on. Everything else is someone else's average.
Whichever control you pick, the expensive mistakes happen one level up: how much to put behind which campaign in the first place, and whether today's spend is on track. That layer is what I built Adscalr's budget planning around. It drafts 3 to 5 prioritized campaign plans from your real 12-week performance, labels every CPI figure as the AI estimate it is, and checks pacing every 5 minutes with staged alerts: runaway at 150% of the cap, overspend at 110%, underspend below 70% after midday. CBO or ABO, you find out before the daily budget is gone. The full mechanism is on the budget intelligence page.
This is the thinking behind Adscalr.
See the product →